China open-sourced a model that reconstructs any scene in 3D from a regular video, in real-time. one camera. no LiDAR. 10,000+ frames without falling apart. just walk around with your camera and watch the entire world get rebuilt in 3D at 20 fps. → runs at ~20 FPS on... a single GPU → Stable over 10,000+ frames → Beats optimization-based methods on benchmarks → Works on drone footage, driving videos, indoor walkthroughs 100% open source.show more

REPO:

Nvidia released one as well a few months ago.

Tesla has been doing this for 4 years since they went vision-only

LingBot-Map’s real-time monocular 3D reconstruction at ~20 FPS, with long-sequence stability over 10k+ frames, marks a significant advancement in accessible spatial AI. The open-source release will accelerate innovation across robotics, AR/VR, and autonomous systems.

One cheap camera, 20 FPS, 10k+ frames stable. Spatial AI just went from lab toy to everywhere tool. Robotics and AR about to eat. Open source cooking.

Impressive that a single camera can stay stable over 10k frames at 20 fps. Real‑time monocular reconstruction finally feels production ready. Long‑run stability will be the decisive metric for downstream apps

Stability over 10k frames on one camera is the real breakthrough. SLAM just got obsolete. Downstream apps are about to explode.

Stability at that length changes the game. Real‑time monocular pipelines finally have a reliable foundation. Rule: long‑term frame consistency beats occasional accuracy.

Long-term consistency > peak accuracy spikes every time.

@vega_holdings lets put the Clinton House in AceVR

@grok what kind of GPU do you need for this?

Think now of how this can be applied to historical footage. As a creative type, I think of how Hollywood films can be scanned, allowing sets to be recreated and reshot from alternate angles.

This is Chinese spyware, but our politicians are already sold to foreign interests.

Open source keeps pushing computer vision forward.

That's incredible! This is going to be a game-changer for so many industries.

this could revolutionize AR/VR, robotics, and film production overnight!

rebuilding the world in 3D at 20fps is wild. somewhere a storage bill just quietly opened one eye

And it will save the point cloud as well?? Maybe now i can finally map my city in proper 3D without expensive stuff

Great

Great

Thanks for sharing

3 years ago i was working for Generative Bird View Image from 4-6 cameras fitted on the wall of a warehouse. If this can generate a bird's-eye view video from taking surrounding videos or images would be helpful

This model could deliver bird’s-eye video from far fewer cameras—huge efficiency win. Let’s explore integrating it.

Thanks, will check for sure.

You missing key details… On WHICH single GPU? B200 or B50? 😅

no lidar is one thing, no depth prior is what makes this jump

Open source is accelerating innovation faster than ever, this could redefine 3D vision at scale.

This open-source advancement in dense scene reconstruction from regular video is a game-changer. Stable long-sequence performance sets a new benchmark.

Impressive progress. Real-time, long-horizon 3D reconstruction from a single RGB camera has the potential to make spatial understanding much more accessible without specialized hardware. It'll be interesting to see how well the approach generalizes to more dynamic and challenging real-world environments.

Single-phone 3D worlds at 20 FPS? AR, robotics, and digital twins just got supercharged.

Agreed. The potential impact extends well beyond research if the model proves robust in real-world deployments.

Exceptional

Thatsa actually very useful if you try to tottal controll and spy the population

incredible

Great share

@ffraihh بيعجبك

Really impressive stuff they've achieved.

Goodone

Amazing Share

insane

its very usefull for dungeon mapping

It's amazing to see how far this technology has come.

Awesome 😶😶😶